{"title":"Attention-based Stacked Supervised Poisson Autoencoders for Defects Prediction in Casting-rolling process","authors":"Xinmin Zhang, Zhihuan Song, Zhiqiang Ge","doi":"10.1109/DDCLS55054.2022.9858528","DOIUrl":null,"url":null,"abstract":"In this paper, we propose an attention-based stacked supervised Poisson autoencoders (ASSPAE) modeling framework to predict the occurrence of defects in the casting-rolling process. ASSPAE is a novel deep learning model, which is designed based on the stacked supervised Poisson autoencoders and attention-based multi-feature fusion strategy. Stacked supervised Poisson autoencoders are a novel deep neural networks model, which is designed by integrating the Poisson regression network layer into the deep autoencoders framework. Compared with the conventional autoencoder-based models, the proposed stacked supervised Poisson autoencoders not only can capture the quality-relevant deep features of the data, but also is suitable for predicting the count-type quality variable. Furthermore, in order to efficiently leverage information present in different layers of the network, an attention-based multi-feature fusion strategy is integrated into the stacked supervised Poisson autoencoders modeling framework. The effectiveness of the proposed ASSPAE method is evaluated by its application to the cast-rolling process of a steelmaking plant. The application results show that the prediction accuracy of the proposed ASSPAE modeling method was improved by 32%, 31%, 24%, and 13% in Root Mean Square Error (RMSE) compared with the partial least square regression, support vector regression, Poisson regression, and stacked autoencoders-Poisson regression, respectively.","PeriodicalId":412672,"journal":{"name":"2022 IEEE 11th Data Driven Control and Learning Systems Conference (DDCLS)","volume":"14 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2022-08-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"1","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2022 IEEE 11th Data Driven Control and Learning Systems Conference (DDCLS)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/DDCLS55054.2022.9858528","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 1
Abstract
In this paper, we propose an attention-based stacked supervised Poisson autoencoders (ASSPAE) modeling framework to predict the occurrence of defects in the casting-rolling process. ASSPAE is a novel deep learning model, which is designed based on the stacked supervised Poisson autoencoders and attention-based multi-feature fusion strategy. Stacked supervised Poisson autoencoders are a novel deep neural networks model, which is designed by integrating the Poisson regression network layer into the deep autoencoders framework. Compared with the conventional autoencoder-based models, the proposed stacked supervised Poisson autoencoders not only can capture the quality-relevant deep features of the data, but also is suitable for predicting the count-type quality variable. Furthermore, in order to efficiently leverage information present in different layers of the network, an attention-based multi-feature fusion strategy is integrated into the stacked supervised Poisson autoencoders modeling framework. The effectiveness of the proposed ASSPAE method is evaluated by its application to the cast-rolling process of a steelmaking plant. The application results show that the prediction accuracy of the proposed ASSPAE modeling method was improved by 32%, 31%, 24%, and 13% in Root Mean Square Error (RMSE) compared with the partial least square regression, support vector regression, Poisson regression, and stacked autoencoders-Poisson regression, respectively.